CN108804499A - A kind of trademark image retrieval method - Google Patents
A kind of trademark image retrieval method Download PDFInfo
- Publication number
- CN108804499A CN108804499A CN201810298625.9A CN201810298625A CN108804499A CN 108804499 A CN108804499 A CN 108804499A CN 201810298625 A CN201810298625 A CN 201810298625A CN 108804499 A CN108804499 A CN 108804499A
- Authority
- CN
- China
- Prior art keywords
- image
- window
- retrieved
- similar
- contrast images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to field of image search more particularly to a kind of trademark image retrieval methods, include the following steps：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction；By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress；Screening correctly matching；Candidate similar area segmentation；Similarity matching between local scale in region, the sequence of retrieval result Similar contrasts' image.Retrieval accuracy of the present invention is high, and omission factor is low, and speed is fast.
Description
Technical field
The invention belongs to field of image search more particularly to a kind of trademark image retrieval methods.
Background technology
Trade mark plays very important effect in industry and commerce society, is the mark of company, product or service, with enterprise
Commercial quality, service quality, management combine together, become the symbol of goodwill, are a kind of intangible assets.
Trade mark retrieval refers to the process of that image similar with input trademark image is found from existing trademark image library.
Although the research of the trademark image retrieval based on content is more active, there are also systems to have put into use,
Still there are some problems not solved preferably.Its maximum difficulty is exactly：The image bottom content characteristic that system extracts
It can not be mapped between the high-level semantic used when user search, that is to say, that characteristics of image is beyond expression user's at all
High-level semantic, therefore content-based retrieval result is often unsatisfactory.
During pervious manual coding, brand authentication personnel can add coding result according to certain rule
Power, originally the weights in trade mark corresponding to part and parcel are big, can protrude the important figure of these in original trademark in this way, and show
The global characteristics of some systems usually abstract image are matched, and cannot protrude these important informations, therefore system is reliable
Property is not high；Another important problem is exactly the speed issue retrieved, with the increase of amount of images in library, the speed of retrieval at
For a bottleneck of system for restricting, and the existing registered trademark accumulated quantity in China oneself through being more than 10,000,000.
Invention content
The present invention provides a kind of trademark image retrieval method, and retrieval accuracy is high, and omission factor is low, and speed is fast.
A kind of trademark image retrieval method, includes the following steps：
S1：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction；
S2：By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress；
S3：Screening correctly matching；
S4：Candidate similar area segmentation；
S5：Similarity matching between local scale in region carries out the sequence of contrast images according to the size of matching similitude.
Preferably, step S1 is divided the region of image to be retrieved and contrast images using the sliding window of multiple and different scales, carried
Take sliding window image in window feature.
Preferably, step S1 includes the following steps：
(1) the gradient orientation histogram feature of sliding window image in window pixel is extracted；
(2) gradient orientation histogram quantization encoding；
(3) it normalizes；
(4) spatial distribution describes, and cascades gradient orientation histogram and spatial distribution.
It is further preferred that step (1), calculates the horizontal gradient and vertical gradient of image slices vegetarian refreshments, using direction template
[- 1,0,1], computation rule are [Gh,Gv]=gradient (F),
It is further preferred that step (2), orientation angle θ=arctan (G of image slices vegetarian refreshmentsv/Gh), it is arrived in plane 0
360 degree are carried out n direction quantizations and one gradient direction are quantized in its adjacent both direction, i.e., using Fuzzy Quantifying
By a direction projection to the representation in components in two neighboring direction, the quantity of the quantized directions of each pixel is counted, is carried out straight
Side's figure statistics, is indicated in the form of one-dimensional matrix.The n values, for the integer more than 2；The numerical value of n is bigger, statistics with histogram
More accurate, the calculation amount of data is also bigger.Preferably, n=6 or 8 or 9 or 12.
Traditional direction quantization method is excessively harsh, and the feature robustness after causing gradient direction to quantify is poor, to direction
It is more sensitive.The present invention uses Fuzzy Quantifying, and the feature robustness after quantization is preferable, and accuracy is high.
It is further preferred that step (3), counted out the method that normalization is combined with area normalization using gradient, ladder
Spend direction histogram Hist=[h0,h1,…,hi]T, sliding window area is pArea, normalized histogramWherein i=n-1, α=1/n.
Gradient is counted out normalized method, so that feature is had good consistency of scale, while embodying each gradient
Direction relative statistic distributed intelligence；The disadvantage is that the variation that some bin gradient is counted out will influence the relative statistic of whole histogram
Distribution.
Area parameters will make feature have relatively good consistency of scale by area evolution to calculate, and be joined based on area
Several histogram method for normalizing, had not only contained the abundant degree of marginal information in characteristic window, but also can reflect each gradient side
To statistical distribution information, the variation of single bin does not interfere with the value of other bin；The disadvantage is that the otherness between each bin may
It reduces, for the window that edge is abundant, the value of each bin is relatively large, there are multiple higher values, and it is diluter for edge
The value of thin window, each bin is smaller, and there are multiple smaller values.
Gradient is counted out the method that normalization is combined with area normalization, both ensured between each bin relatively solely
Vertical property, and take into account the otherness of each bin statistical distributions.
It is further preferred that step (4), each region of image is subjected to quantization encoding, all directions gradient in statistical picture
The position of centre of gravity of point, the region of image is fallen into according to position of centre of gravity, using the position encoded of the region, in gradient orientation histogram
It is cascaded after eigenmatrix position encoded.
Gradient orientation histogram and spatial distribution description are combined, are conducive to improve accuracy when Image Feature Matching；
Each region of image is carried out to the spatial distribution describing mode of quantization encoding, position definition is accurate, and data volume is small, and calculating speed is fast.
Preferably, step S2 slides sliding window in image to be retrieved, traverses and all in contrast images meets similar possibility
Similarity distance is calculated in window.
It is further preferred that step S2, the mode of sliding window is slided in image to be retrieved, for from the center of image to be retrieved to
Surrounding is slided.
It is further preferred that step S2, similarity distance is calculated by Hamming distance (Hamming Distance)；Phase
Like distanceWherein, feature binary string of the image to be retrieved after coding is fi, contrast images process
Feature binary string after coding is gj, fi kIndicate binary string fiKth position, gj kIndicate binary string gjKth position,Table
Show that xor operation, the value of α are equal to feature binary string fiWith gjThe inverse of length sum.
It is further preferred that step S2, meeting the condition that similar possibility needs meet is：(1) contrast images window
Center near image sliding window window center position to be retrieved, permission transformation range is u, and the value range of u is 0.4
To 0.6；(2) contrast images window has similar length-width ratio, the ratio of described two length-width ratios with image sliding window window to be retrieved
Value ranging from 0.2 to 5, preferably 0.5 to 2.
Preferably, step S3 eliminates erroneous matching using the method for scale-space consistency；It is consistent using random sampling
Property (RANSAC) algorithm, be retained on scale and on spatial position all it is consistent matching pair, the matching pair of debug.
It is further preferred that step S3, specific algorithm are：If a pair of of match window of image to be retrieved and contrast images
{(x1,y1),(x1′,y1′)}:{(x2,y2),(x2′,y2') (wherein, (x1,y1)、(x1′,y1') image to be retrieved is indicated respectively
The upper left corner of window and bottom right angular coordinate, (x2,y2)、(x2′,y2') respectively indicate contrast images window the upper left corner and the lower right corner
Coordinate), then there are space transform modelsSo that L can be solved,
Wherein ɑ 1, ɑ 2 are the relevant zooming parameter of specific matching window, and tx, ty are and the relevant translation parameters of specific matching window；It is right
Space transform models L uses random sampling consistency (RANSAC) algorithm, is retained on scale and on spatial position all with one
The matching pair of cause property, the matching pair of debug.
Preferably, step S4 goes out similar area according to adaptive threshold fuzziness；Correct match window is carried out quantitative
Weighted superposition counts the number of each similar window, according to adaptive threshold Factorization algorithm similar area.
It is further preferred that step S4, structure positioning point (anchor is defined as by the center of step S1 sliding windows
Point), in step S4, statistics covers the number of the similar window of each structure positioning point (anchor point).
More similar region, the number for covering the similar window of the regional structure anchor point (anchor point) are more.
It is the weight of each pair of match window by similarity distance d it is further preferred that step S4ijIt determines, similarity distance is got over
Small, the weight given is bigger, and similarity distance is bigger, and the weight given is smaller, and overall average weight is 1.
It is further preferred that step S4, if T0The gross area for initial threshold matrix, all similar windows is s, then adaptive
The threshold matrix T=κ T answered0(s/(100))α, wherein κ, α is experience numerical constant, as the running parameter of sliding window specification is answered
Carry out the adjustment of adaptability.
Preferably, step S5, the similitude of each contrast images similar area, passes through Hamming distance (Hamming
Distance it) is calculated；Similar Window match in region, lookup method are searched for local neighborhood.
It is further preferred that step S5 passes through the arbitrary sliding of sliding window, traversal pair in the similar area of image to be retrieved
Than in the similar area of image, all sliding window windows for meeting similar possible condition, are calculated similarity distance, similarity distance
Minimum is most like window.
It is further preferred that the similitude of sliding window is with sliding window in step S5, image to be retrieved and contrast images similar area
The similitude of structure positioning point (anchor point) replace, similarity distance is by all with the structure positioning point (anchor
Point the mean value of the correspondence similarity distance of the window centered on) calculates.
The similarity distance d of image similar area to be retrieved and contrast images similar areaAB, specific algorithm is：Wherein, nAFor in image similar area to be retrieved include structure positioning point (anchor
Point number), nBTo include the number of structure positioning point (anchor point) in contrast images similar area, (u, v) is
The coordinate of structure positioning point (anchor point), dAUVFor image similar area structure positioning point (u, v) to be retrieved it is similar away from
From dBUVFor the similarity distance of contrast images similar area structure positioning point (u, v), λ is similar area parameters and nA、nBAt anti-
Than the similar area gross area is bigger, and λ is smaller.
It is further preferred that step S5, the similarity distance of image more to be retrieved and all contrast images in image library, root
The sequence of retrieval result Similar contrasts' image is carried out according to similarity distance.Similarity distance is smaller, and image to be retrieved is got over contrast images
Similar, sequence is more forward.
Advantageous effect：
1, the present invention uses Fuzzy Quantifying, and the feature robustness after quantization is preferable, and accuracy is high.
2, the present invention counts out gradient the method that normalization is combined with area normalization, has both ensured between each bin
Relative independentability, and take into account the otherness of each bin statistical distributions.
3, by gradient orientation histogram and spatial distribution description combine, be conducive to improve Image Feature Matching when it is accurate
Property；Each region of image is carried out to the spatial distribution describing mode of quantization encoding, position definition is accurate, and data volume is small, calculating speed
Soon.
4, the method that the present invention uses scale-space consistency increases compared to traditional simple consistency of scale method
Corresponding relationship, combines the feature in graphical rule and space, energy while having added the correspondence and scale-space in space
The matching of mistake is effectively eliminated, fallout ratio is reduced, accuracy is high.
5, the present invention goes out similar area according to adaptive threshold fuzziness, and by way of weighted superposition, retrieval result has more
There is accuracy.
6, retrieval result of the invention is ranked up according to similarity distance, scientific and reasonable, and the judgement with people is close to unanimously.
Specific implementation mode
The technical solution in the embodiment of the present invention will be clearly and completely described below, it is clear that described reality
It is only a part of the embodiment of the present invention to apply example, instead of all the embodiments.Based on the embodiments of the present invention, this field is general
The every other embodiment that logical technical staff is obtained without making creative work belongs to what the present invention protected
Range.
Embodiment 1
A kind of trademark image retrieval method, includes the following steps：
S1：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction；
S2：By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress；
S3：Screening correctly matching；
S4：Candidate similar area segmentation；
S5：Similarity matching between local scale in region carries out the sequence of contrast images according to the size of matching similitude.
Embodiment 2
A kind of trademark image retrieval method, includes the following steps：
S1：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction；Using multiple and different rulers
The sliding window of degree divides the region of image to be retrieved and contrast images, extracts sliding window image in window feature；
Include the following steps：
(1) the gradient orientation histogram feature of sliding window image in window pixel is extracted；
(2) gradient orientation histogram quantization encoding；
(3) it normalizes；
(4) spatial distribution describes, and cascades gradient orientation histogram and spatial distribution.
S2：By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress；In image to be retrieved
Sliding window is slided, all windows for meeting similar possibility in contrast images is traversed, similarity distance is calculated；In image to be retrieved
The mode for sliding sliding window, to be slided around from the center of image to be retrieved；
Similarity distance is calculated by Hamming distance (Hamming Distance)；Similarity distanceWherein, feature binary string of the image to be retrieved after coding is fi, contrast images are after coding
Feature binary string be gj, fi kIndicate binary string fiKth position, gj kIndicate binary string gjKth position,Indicate exclusive or
The value of operation, α is equal to feature binary string fiWith gjThe inverse of length sum；
Meeting the condition that similar possibility needs meet is：(1) center of contrast images window is in figure to be retrieved
Near sliding window window center position, permission transformation range is u, and the value range of u is 0.4 to 0.6；(2) contrast images window
Mouth has similar length-width ratio with image sliding window window to be retrieved, and the ratio range of described two length-width ratios is 0.2 to 5, preferably
It is 0.5 to 2.
S3：Screening correctly matching；Erroneous matching is eliminated using the method for scale-space consistency；Using random sampling
Consistency (RANSAC) algorithm, be retained on scale and on spatial position all it is consistent matching pair, of debug
Pairing.
S4：Candidate similar area segmentation.
S5：Similarity matching between local scale in region carries out the sequence of contrast images according to the size of matching similitude；Respectively
The similitude of contrast images similar area is calculated by Hamming distance (Hamming Distance)；Similar window in region
Mouth matching, lookup method are that local neighborhood is searched.
Embodiment 3
A kind of trademark image retrieval method, includes the following steps：
S1：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction；Using multiple and different rulers
The sliding window of degree divides the region of image to be retrieved and contrast images, extracts sliding window image in window feature；
Include the following steps：
(1) the gradient orientation histogram feature of sliding window image in window pixel is extracted；
(2) gradient orientation histogram quantization encoding；
(3) it normalizes；
(4) spatial distribution describes, and cascades gradient orientation histogram and spatial distribution；
Step (1) calculates the horizontal gradient and vertical gradient of image slices vegetarian refreshments, using direction template [- 1,0,1], calculates
Rule is [Gh,Gv]=gradient (F),
Step (2), orientation angle θ=arctan (G of image slices vegetarian refreshmentsv/Gh), carry out 8 directions with 0 to 360 degree in plane
Quantization, using Fuzzy Quantifying, a gradient direction is quantized in its adjacent both direction, i.e., throws a direction
Shadow counts the quantity of the quantized directions of each pixel to the representation in components in two neighboring direction, statistics with histogram is carried out, using one
The form for tieing up matrix indicates；
Step (3) is counted out the method that normalization is combined with area normalization using gradient, gradient orientation histogram
Hist=[h0,h1,…,h7]T, sliding window area is pArea, normalized histogramIts
Middle i=7, α=1/8；
Each region of image is carried out quantization encoding by step (4), the position of centre of gravity of all directions gradient point in statistical picture,
The region that image is fallen into according to position of centre of gravity, using the position encoded of the region, in gradient orientation histogram eigenmatrix rear class
Join position encoded.For example, sliding window is divided into 3 multiply 39 pieces, each region allocated code is as follows：
| 1001 | 1000 | 1100 |
| 0001 | 0000 | 0100 |
| 0011 | 0010 | 0110 |
The position of centre of gravity of all directions gradient point in statistical picture, position of centre of gravity fall into some region of image, such as heavy
Heart position falls into the upper left hand corner section in sliding window region, then uses position encoded the 1001 of the region, in gradient orientation histogram spy
Position encoded 1001 are cascaded after sign matrix.
S2：By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress.
S3：Screening correctly matching；Erroneous matching is eliminated using the method for scale-space consistency；Using random sampling
Consistency (RANSAC) algorithm, be retained on scale and on spatial position all it is consistent matching pair, of debug
Pairing；
Specific algorithm is：If a pair of of match window of image to be retrieved and contrast images
Wherein, (x1,y1)、(x1′,y1') upper left corner and the bottom right angular coordinate of image window to be retrieved, (x are indicated respectively2,y2)、(x2′,
y2') respectively indicate contrast images window the upper left corner and bottom right angular coordinate), then there are space transform models
So that L can be solved, wherein ɑ 1, ɑ 2 are the relevant scaling ginseng of specific matching window
Number, tx, ty are and the relevant translation parameters of specific matching window；Random sampling consistency is used to space transform models L
(RANSAC) algorithm, be retained on scale and on spatial position all it is consistent matching pair, the matching pair of debug.
S4：Candidate similar area segmentation.
S5：Similarity matching between local scale in region carries out the sequence of contrast images according to the size of matching similitude.
Embodiment 4
A kind of trademark image retrieval method, includes the following steps：
S1：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction.
S2：By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress.
S3：Screening correctly matching.
S4：Candidate similar area segmentation.
S5：Similarity matching between local scale in region carries out the sequence of contrast images according to the size of matching similitude；Respectively
The similitude of contrast images similar area is calculated by Hamming distance (Hamming Distance)；Similar window in region
Mouth matching, lookup method are that local neighborhood is searched；
In the similar area of image to be retrieved, by the arbitrary sliding of sliding window, institute in the similar area of contrast images is traversed
There is the sliding window window for meeting similar possible condition, be calculated similarity distance, similarity distance minimum is most like window；
The similitude of sliding window is with the structure positioning point (anchor of sliding window in image to be retrieved and contrast images similar area
Point similitude) replaces, and similarity distance is by all windows centered on the structure positioning point (anchor point)
The mean value of similarity distance is corresponded to calculate；
The similarity distance d of image similar area to be retrieved and contrast images similar areaAB, specific algorithm is：Wherein, nAFor in image similar area to be retrieved include structure positioning point (anchor
Point number), nBTo include the number of structure positioning point (anchor point) in contrast images similar area, (u, v) is
The coordinate of structure positioning point (anchor point), dAUVFor image similar area structure positioning point (u, v) to be retrieved it is similar away from
From dBUVFor the similarity distance of contrast images similar area structure positioning point (u, v), λ is similar area parameters and nA、nBAt anti-
Than the similar area gross area is bigger, and λ is smaller；
The similarity distance of image more to be retrieved and all contrast images in image library, retrieval knot is carried out according to similarity distance
The sequence of fruit Similar contrasts' image.Similarity distance is smaller, and image to be retrieved is more similar to contrast images, and sequence is more forward.
Embodiment 5
A kind of trademark image retrieval method, includes the following steps：
S1：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction.
S2：By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress.
S3：Screening correctly matching.
S4：Candidate similar area segmentation；Go out similar area according to adaptive threshold fuzziness；By correct match window into line number
Weighted superposition in amount counts the number of each similar window, according to adaptive threshold Factorization algorithm similar area；
The center of step S1 sliding windows is defined as structure positioning point (anchor point), in step S4, statistics is covered
Cover the number of the similar window of each structure positioning point (anchor point)；
For each pair of match window weight by similarity distance dijIt determines, similarity distance is smaller, and the weight given is bigger, similar
Distance is bigger, and the weight given is smaller, and overall average weight is 1；
If T0The gross area for initial threshold matrix, all similar windows is s, then adaptive threshold matrix T=κ T0(s/
(100))α, wherein κ, α is experience numerical constant, and the adjustment of adaptability should be carried out with the running parameter of sliding window specification.
S5：Similarity matching between local scale in region carries out the sequence of contrast images according to the size of matching similitude.
Finally it should be noted that：The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention,
Although the present invention is described in detail referring to the foregoing embodiments, for those skilled in the art, still may be used
With technical scheme described in the above embodiments is modified or equivalent replacement of some of the technical features,
All within the spirits and principles of the present invention, any modification, equivalent replacement, improvement and so on should be included in the present invention's
Within protection domain.
Claims (10)
1. a kind of trademark image retrieval method, which is characterized in that include the following steps：
S1：Contrast images in image to be retrieved and image library are subjected to Multi resolution feature extraction；
S2：By Similarity matching between characteristics of image to be retrieved and the global scale of contrast images feature progress；
S3：Screening correctly matching；
S4：Candidate similar area segmentation；
S5：Similarity matching between local scale in region carries out the sequence of contrast images according to the size of matching similitude.
2. trademark image retrieval method according to claim 1, which is characterized in that the step S1, use are multiple and different
The sliding window of scale divides the region of image to be retrieved and contrast images, extracts sliding window image in window feature；
The step S1, includes the following steps：
(1) the gradient orientation histogram feature of sliding window image in window pixel is extracted；
(2) gradient orientation histogram quantization encoding；
(3) it normalizes；
(4) spatial distribution describes, and cascades gradient orientation histogram and spatial distribution.
3. trademark image retrieval method according to claim 2, which is characterized in that the step (1) calculates image pixel
The horizontal gradient and vertical gradient of point, using direction template [- 1,0,1], computation rule is [Gh,Gv]=gradient (F),
The step (2), orientation angle θ=arctan (G of image slices vegetarian refreshmentsv/Gh), carry out the directions n with 0 to 360 degree in plane
Quantization, using Fuzzy Quantifying, a gradient direction is quantized in its adjacent both direction, i.e., throws a direction
Shadow counts the quantity of the quantized directions of each pixel to the representation in components in two neighboring direction, statistics with histogram is carried out, using one
The form for tieing up matrix indicates；The n values, for the integer more than 2；Preferably, n=6 or 8 or 9 or 12；
The step (3) is counted out the method that normalization is combined with area normalization using gradient, gradient orientation histogram
Hist=[h0,h1,…,hi]T, sliding window area is pArea, normalized histogramWherein
I=n-1, α=1/n；
Each region of image is carried out quantization encoding by the step (4), the position of centre of gravity of all directions gradient point in statistical picture,
The region that image is fallen into according to position of centre of gravity, using the position encoded of the region, in gradient orientation histogram eigenmatrix rear class
Join position encoded.
4. trademark image retrieval method according to claim 2, which is characterized in that the step S2, in image to be retrieved
Sliding window is slided, all windows for meeting similar possibility in contrast images is traversed, similarity distance is calculated；The step S2,
The mode that sliding window is slided in image to be retrieved, to be slided around from the center of image to be retrieved.
5. trademark image retrieval method according to claim 4, which is characterized in that the step S2, similarity distance pass through
Hamming distance (Hamming Distance) is calculated；Similarity distanceWherein, image warp to be retrieved
The feature binary string crossed after coding is fi, feature binary string of the contrast images after coding is gj, fi kIndicate binary system
String fiKth position, gj kIndicate binary string gjKth position,Indicate that xor operation, the value of α are equal to feature binary string fiWith
gjThe inverse of length sum；
The step S2, meeting the condition that similar possibility needs meet is：(1) center of contrast images window is being waited for
It retrieves near image sliding window window center position, permission transformation range is u, and the value range of u is 0.4 to 0.6；(2) it compares
Image window with image sliding window window to be retrieved there is similar length-width ratio, the ratio range of described two length-width ratios to be arrived for 0.2
5, preferably 0.5 to 2.
6. trademark image retrieval method according to claim 4, which is characterized in that the step S3, using scale-space
The method of consistency eliminates erroneous matching；Using random sampling consistency (RANSAC) algorithm, it is retained on scale and space bit
Set all consistent matching pair, the matching pair of debug.
7. trademark image retrieval method according to claim 4, which is characterized in that the step S4, according to adaptive thresholding
Value is partitioned into similar area；Correct match window is subjected to quantitative weighted superposition, counts the number of each similar window, root
According to adaptive threshold Factorization algorithm similar area.
8. trademark image retrieval method according to claim 7, which is characterized in that the step S4, by step S1 sliding windows
Center be defined as structure positioning point (anchor point), in step S4, statistics covers each structure positioning point
The number of the similar window of (anchor point)；The step S4 is the weight of each pair of match window by similarity distance dijCertainly
Fixed, similarity distance is smaller, and the weight given is bigger, and similarity distance is bigger, and the weight given is smaller, and overall average weight is 1.
9. according to claim 1 to 8 any one of them trademark image retrieval method, which is characterized in that the step S5 is each right
Than the similitude of image similar area, calculated by Hamming distance (Hamming Distance)；Similar window in region
Matching, lookup method are searched for local neighborhood；
The step S5 traverses the similar area of contrast images in the similar area of image to be retrieved by the arbitrary sliding of sliding window
Similarity distance is calculated in all sliding window windows for meeting similar possible condition in domain, and similarity distance minimum is most like
Window.
10. trademark image retrieval method according to claim 9, which is characterized in that the step S5, image to be retrieved with
The similitude of sliding window is replaced with the similitude of the structure positioning point (anchor point) of sliding window in contrast images similar area,
Similarity distance by the mean value of the correspondence similarity distance of all windows centered on the structure positioning point (anchor point) Lai
It calculates；
The similarity distance d of image similar area to be retrieved and contrast images similar areaAB, specific algorithm is：Wherein, nAFor in image similar area to be retrieved include structure positioning point (anchor
Point number), nBTo include the number of structure positioning point (anchor point) in contrast images similar area, (u, v) is
The coordinate of structure positioning point (anchor point), dAUVFor image similar area structure positioning point (u, v) to be retrieved it is similar away from
From dBUVFor the similarity distance of contrast images similar area structure positioning point (u, v), λ is similar area parameters and nA、nBAt anti-
Than the similar area gross area is bigger, and λ is smaller.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810298625.9A CN108804499B (en) | 2018-04-03 | 2018-04-03 | Trademark image retrieval method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810298625.9A CN108804499B (en) | 2018-04-03 | 2018-04-03 | Trademark image retrieval method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108804499A true CN108804499A (en) | 2018-11-13 |
| CN108804499B CN108804499B (en) | 2022-04-29 |
Family
ID=64095323
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810298625.9A Active CN108804499B (en) | 2018-04-03 | 2018-04-03 | Trademark image retrieval method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108804499B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117788850A (en) * | 2024-02-21 | 2024-03-29 | 深圳欧税通技术有限公司 | Trademark similarity evaluation method and device |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060274081A1 (en) * | 2005-06-03 | 2006-12-07 | Canon Kabushiki Kaisha | Image retrieving apparatus, image retrieving method, program, and storage medium |
| CN101576956A (en) * | 2009-05-11 | 2009-11-11 | 天津普达软件技术有限公司 | On-line character detection method based on machine vision and system thereof |
| US20100290708A1 (en) * | 2009-05-12 | 2010-11-18 | Canon Kabushiki Kaisha | Image retrieval apparatus, control method for the same, and storage medium |
| CN102541954A (en) * | 2010-12-29 | 2012-07-04 | 北京大学 | Method and system for searching trademarks |
| US20130208997A1 (en) * | 2010-11-02 | 2013-08-15 | Zte Corporation | Method and Apparatus for Combining Panoramic Image |
| CN104462380A (en) * | 2014-12-11 | 2015-03-25 | 北京中细软移动互联科技有限公司 | Trademark retrieval method |
| CN104462381A (en) * | 2014-12-11 | 2015-03-25 | 北京中细软移动互联科技有限公司 | Trademark image retrieval method |
| CN105139023A (en) * | 2015-07-24 | 2015-12-09 | 福州大学 | Seed identification method based on multi-scale feature fusion and extreme learning machine |
| CN105160654A (en) * | 2015-07-09 | 2015-12-16 | 浙江工商大学 | Towel label defect detecting method based on feature point extraction |
| CN105550381A (en) * | 2016-03-17 | 2016-05-04 | 北京工业大学 | Efficient image retrieval method based on improved SIFT (scale invariant feature transform) feature |
-
2018
- 2018-04-03 CN CN201810298625.9A patent/CN108804499B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060274081A1 (en) * | 2005-06-03 | 2006-12-07 | Canon Kabushiki Kaisha | Image retrieving apparatus, image retrieving method, program, and storage medium |
| CN101576956A (en) * | 2009-05-11 | 2009-11-11 | 天津普达软件技术有限公司 | On-line character detection method based on machine vision and system thereof |
| US20100290708A1 (en) * | 2009-05-12 | 2010-11-18 | Canon Kabushiki Kaisha | Image retrieval apparatus, control method for the same, and storage medium |
| US20130208997A1 (en) * | 2010-11-02 | 2013-08-15 | Zte Corporation | Method and Apparatus for Combining Panoramic Image |
| CN102541954A (en) * | 2010-12-29 | 2012-07-04 | 北京大学 | Method and system for searching trademarks |
| CN104462380A (en) * | 2014-12-11 | 2015-03-25 | 北京中细软移动互联科技有限公司 | Trademark retrieval method |
| CN104462381A (en) * | 2014-12-11 | 2015-03-25 | 北京中细软移动互联科技有限公司 | Trademark image retrieval method |
| CN105160654A (en) * | 2015-07-09 | 2015-12-16 | 浙江工商大学 | Towel label defect detecting method based on feature point extraction |
| CN105139023A (en) * | 2015-07-24 | 2015-12-09 | 福州大学 | Seed identification method based on multi-scale feature fusion and extreme learning machine |
| CN105550381A (en) * | 2016-03-17 | 2016-05-04 | 北京工业大学 | Efficient image retrieval method based on improved SIFT (scale invariant feature transform) feature |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117788850A (en) * | 2024-02-21 | 2024-03-29 | 深圳欧税通技术有限公司 | Trademark similarity evaluation method and device |
| CN117788850B (en) * | 2024-02-21 | 2024-05-10 | 深圳欧税通技术有限公司 | Trademark similarity evaluation method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108804499B (en) | 2022-04-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11163818B2 (en) | Media fingerprinting and identification system | |
| Shubham et al. | A generalized Masi entropy based efficient multilevel thresholding method for color image segmentation | |
| CN109615014A (en) | A kind of data sorting system and method based on the optimization of KL divergence | |
| Guedri et al. | Indexing and images retrieval by content | |
| CN104850822B (en) | Leaf identification method under simple background based on multi-feature fusion | |
| CN108845998B (en) | Trademark image retrieval and matching method | |
| Yan et al. | Adaptive fusion of color and spatial features for noise-robust retrieval of colored logo and trademark images | |
| CN110188625B (en) | Video fine structuring method based on multi-feature fusion | |
| Cheng et al. | DRNet: Towards fast, accurate and practical dish recognition | |
| Le et al. | Geodesic distance histogram feature for video segmentation | |
| Belalia et al. | Region-based image retrieval in the compressed domain using shape-adaptive DCT | |
| CN108694411B (en) | Method for identifying similar images | |
| CN108764245B (en) | Method for improving similarity judgment accuracy of trademark graphs | |
| CN104732246B (en) | A kind of semi-supervised coorinated training hyperspectral image classification method | |
| CN109902692A (en) | A kind of image classification method based on regional area depth characteristic coding | |
| CN108804499A (en) | A kind of trademark image retrieval method | |
| CN108763265A (en) | A kind of image-recognizing method based on block research | |
| Xu et al. | Fast Feature Extraction Based on Multi-feature Classification for Color Image. | |
| Fu et al. | Pedestrian detection by feature selected self-similarity features | |
| Das et al. | Multi technique amalgamation for enhanced information identification with content based image data | |
| CN110162654A (en) | It is a kind of that image retrieval algorithm is surveyed based on fusion feature and showing for search result optimization | |
| CN108875753A (en) | A kind of trademark image retrieval feature extracting method | |
| Zhao et al. | Image retrieval based on color features and information entropy | |
| CN108846843A (en) | A kind of image characteristic extracting method | |
| CN108897746A (en) | A kind of image search method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |